Questions about istio external authorization - kubernetes

Problem statement:
My goal is to have istio with external authorization service (ideally HTTP, if not possible than GRPC would do as well). There is a requirement to be able to control what exact status code will be returned to client on authorization service. The latter requirement is the most problematic part.
My research
I have read istio documentation on external authorizer
I have made a prototype with HTTP Auth service, but whatever non 200 status
code I return from Auth Service the client always receives 403
Forbidden
In mesh config specification I see the only possibility to set statusOnError but it will be used only in case auth service is unreachable and it can not be dynamically changed.
Also in envoy documentation for GRPC service I see possibility to set custom status
HTTP attributes for a denied response.
{
"status": "{...}",
"headers": [],
"body": "..."
}
Questions:
Is having custom status possible only with GRPC auth service?
Is istio using envoy API-V3 or API-V2?
Any suggestion how to cook istio with external authorizer and custin status codes?

I made the GRPC Auth service prototype and found the answer. It is counter-intuitive but GRPC external auth service is really more flexible than HTTP one. And it really allows to set arbitrary status code

Related

How to use JWT Auth0 token for Cloud Run Service to Service communication if the Metaserver Token is overriding the Auth0 Token

Prerequisites
I have two Cloud Run services a frontend and a backend. The frontend is written in Vue.js/Nuxt.js and is using a Node backend therefore. The backend is written in Kotlin with Spring Boot.
Problem
To have an authenticated internal communication between the frontend and the backend I need to use a token thttps://cloud.google.com/run/docs/authenticating/service-to-service#javahat is fetched from the google metaserver. This is documented here: https://cloud.google.com/run/docs/authenticating/service-to-service#java
I did set it all up and it works.
For my second layer of security I integrated the Auth0 authentication provider both in my frontend and my backend. In my frontend a user can log in. The frontend is calling the backend API. Since only authorized users should be able to call the backend I integrated Spring Security to secure the backend API endpoints.
Now the backend verifies if the token of the caller's request are valid before allowing it to pass on to the API logic.
However this theory does not work. And that simply is because I delegate the API calls through the Node backend proxy. The proxy logic however is already applying a token to the request to the backend; it is the google metaserver token. So let me illustrate that:
Client (Browser) -> API Request with Auth0 Token -> Frontend Backend Proxy -> Overriding Auth0 Token with Google Metaserver Token -> Calling Backend API
Since the backend is receiving the metaserver token instead of the Auth0 Token it can never successfully authorize the API call.
Question
Due the fact that I was not able to find any articles about this problem I wonder if it's simply because I am doing it basically wrong.
What do I need to do to have a valid Cloud Run Service to Service communication (guaranteed by the metaserver token) but at the same time have a secured backend API with Auth0 authorization?
I see two workarounds to make this happen:
Authorize the API call in the Node backend proxy logic
Make the backend service public available thus the metaserver token is unnecessary
I don't like any of the above - especially the latter one. I would really like to have it working with my current setup but I have no idea how. There is no such thing like multiple authorization token, right?
Ok I figured out a third way to have a de-facto internal service to service communication.
To omit the meta-server token authentication but still restrict access from the internet I did the following for my backend cloud run service:
This makes the service available from the internet however the ingress is preventing any outsider from accessing the service. The service is available without IAM but only for internal traffic.
So my frontend is calling the backend API now via the Node backend proxy. Even though the frontend node-backend and the backend service are both somewhat "in the cloud" they do not share the same "internal network". In fact the frontend node-backend requests would be redirected via egress to the internet and call the backend service just like any other internet-user would do.
To make it work "like it is coming from internal" you have to do something similar like VPN but it's called VPC (Virtual Private Cloud). And luckily that is very simple. Just create a VPC Connector in GCP.
BUT be aware to create a so called Serverless VPC Access (Connector). Explained here: https://cloud.google.com/vpc/docs/serverless-vpc-access
After the Serverless VPC Access has been created you can select it in your Cloud Run Service "Connection" settings. For the backend service it can be simply selected. For the frontend service however it is important to select the second option:
At least that is important in my case since I am calling the backend service by it's assigned service URL instead of a private IP.
After all that is done my JWT token from the frontend is successfully delivered to the backend API without being overwritten by a MetaServer token.

Kubernetes API proxy for HTTPS endpoints

Kubernetes surfaces an API proxy, which allows querying the internal services via eg: https://myhost.com/api/v1/proxy/namespaces/kube-system/services/kubernetes-dashboard/
This is all well, and good. However, for security & compliance reasons, all of our services expose an HTTPS endpoint. Attempting to access them by going to https://myhost/api/v1/proxy/namespaces/default/services/myhttpsservice:3000/ results in
Error: 'read tcp 172.20.122.129:48830->100.96.29.113:3000: read: connection reset by peer'
Trying to reach: 'http://100.96.29.113:3000/'
Because the endpoint, 100.96.29.113:3000 is in fact https.
Is there any way to configure the proxy to apply SSL to specific service endpoints?
(Edit: If this is not currently possible, a relevant github issue link for tracking the feature request is also acceptable answer until it will be)
As documented at https://kubernetes.io/docs/tasks/access-application-cluster/access-cluster/#manually-constructing-apiserver-proxy-urls, (and pointed out on slack), you can access services behind HTTPS by prefixing the servicename with "https:" ;
Using the example from above, correctly it would be: https://myhost/api/v1/proxy/namespaces/default/services/https:myhttpsservice:3000/

How to block HTTP and allows only HTTPS for AWS API Gateway with custom domain name map

I've added certificate with custom domain name map in AWS API gateway but it allows HTTP automatically, how can I block normal HTTP and only allows HTTPS?
All API Gateway APIs are fronted with a CloudFront distribution. Each of these CloudFront distributions (whether it's a Custom Domain like yours or the default *.execute-api distribution) is configured to redirect all HTTP requests to HTTPS. Although CloudFront has the option to strictly require HTTPS and return 403 on HTTP requests we currently don't expose this option for simplicity.
If you feel you have valid use case for requiring HTTPS without a redirect please open a support ticket and the team can evaluate your request.

Generic Paths with AWS API Gateway

Let's say I have a RESTful web service with the following API:
/
/things
/v2
/heartbeat
GET
/stuff
GET
POST
...
This service is running in an AWS Elastic Beanstalk behind an AWS API Gateway. The service itself handles routing beyond the /things path so I don't want to duplicate this routing logic in the API Gateway configuration. Is there a way I can setup my AWS API Gateway to handle any request that is /things and pass that to my service where the service will then handle routing to the correct path and method?
In other words, my API Gateway would handle a request to https://myUrl.com/things/v2/heartbeat and another request to https://myUrl.com/things/v2/stuff using the same /things resource defined in my API Gateway rather than having to define the /heartbeat and /stuff resources in the API Gateway itself.
You can use API Gateway's catch-all path variable to do that.
In your case, you'd have an method of type ANY, with the path /things/{proxy+}
The Endpoint URL you'd use would be something like: https://api.yourbackend.com/{proxy}
Here is a screenshot showing an example HTTP proxy integration in the API Gateway console:
This blog post has more details and screen shots:
https://aws.amazon.com/blogs/aws/api-gateway-update-new-features-simplify-api-development/

AWS API Gateway - Test works, deployed API errors. Why?

I'm trying to setup AWS Api Gateway as a reverse proxy for my actual deployed API.
My understanding is that I do this by creating a "Proxy" Resource and then specifying my http endpoint URL - as described here
Create and Test an API with HTTP Proxy Integration through a Proxy Resource
This works fine when I try to use the API through the "Test" function within the Resource Editor. I can make calls to any exposed resources using GET methods and see the successful responses.
However, when I deploy the API Gateway API I can no longer access anything using the "Invoke URL" it gives me - I simply get:
{
"Message": "No HTTP resource was found that matches the request URI 'http://<myuniqueid>.execute-api.eu-west-1.amazonaws.com/api/Sector/100'.",
"MessageDetail": "No type was found that matches the controller named 'Sector'."
}
If I remove the "Use HTTP Proxy integration" checkbox from the "Integration Request" I can get it working, but why doesn't it work as a proxy?
I suspect that this is caused by a known issue with the HTTP proxy integration. When you use an HTTP proxy integration, API Gateway passes all headers through to the integration endpoint, including the HOST header. Many existing http endpoint require the use of a HOST header which matches their DNS name and in such cases, passing through the HOST header of the API Gateway can confuse the endpoint.
UPDATE: We identified a work-around for this issue.
In your integration request, explicitly add a header named "Host" and give it the value of the integration endpoint DNS name. This will replace the Host header forwarded from the incoming client request with the Host header you specify. This should allow your backend endpoint to function correctly.